Ultralytics assets puts logos, model weights and datasets behind three different fetch paths
Shared Ultralytics logos, media, sample images, pretrained model weights, dataset artifacts, and release assets.
At a glance
- What is it?
- The shared media repository behind the YOLO documentation, app store listings and model downloads. Images are referenced by unpinned raw URLs on the default branch, weights are pulled silently by the Python package, and dataset artifacts arrive through releases, so the same repository distributes three unrelated payload types by three mechanisms.
- Who is it for?
- The repository fits exactly one job, which is giving the Ultralytics ecosystem stable places to point at, and it does that job without becoming a package you install. It does not fit anyone who needs reproducible documentation builds, because every documented image URL tracks the default branch rather than a tag, or anyone treating it as a general asset bucket, since the layout is organised around one product's media.
- Can I use it commercially?
- Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
- Is it still maintained?
- Yes. The repository last received commits 40 days ago.
- What is it written in?
- GitHub does not report a main language for this repository.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- ultralytics/assets702 ★ · +29
| Repository | 2026-09-17 | 2026-11-15 |
|---|---|---|
| ultralytics/assets | 673 | 702 |
29 Stars in the last 59 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
One repository, three unrelated kinds of payload
The description gives it away: shared logos, media, sample images, pretrained model weights, dataset artifacts and release assets in a single place. The page then splits them three ways. Visual assets cover logos, favicons, social icons, Open Graph images, screenshots, diagrams and banners. Model assets are the pretrained files that Ultralytics tools download by themselves when a requested model is not already on disk. Dataset assets are release artifacts and examples that the documentation and tutorials link to. The repository's detected primary language is unknown, which is consistent with the contents: the only code on the page is a three-line Python example, and the rest is media fetched over HTTP. The stated purpose is to give the YOLO ecosystem stable files, covering object detection, instance segmentation, classification, pose estimation and tracking, and the same assets are consumed by documentation, applications, examples, releases and model downloads at once. That breadth is the reason a single repository grew three storage strategies, since git handles a logo, a multi-hundred-megabyte checkpoint and a zip archive very differently.
Documented image URLs all track the default branch
The linking convention is one line long, and it points at `main`:
https://raw.githubusercontent.com/ultralytics/assets/main/path/to/asset.extThe worked example is the logotype SVG at a concrete path:
https://raw.githubusercontent.com/ultralytics/assets/main/logo/Ultralytics_Logotype_Original.svgThat is convenient and unpinned. A documentation page, notebook or marketing site that copies this pattern re-fetches whatever is on the default branch at render time, so an asset can be replaced, moved or restyled without any version bump and without the referring site changing a line. The upside is that a fix to a logo propagates everywhere at once. The cost is that nothing downstream can claim to be showing the file that existed when the page was written, and a broken rename upstream turns into a broken image in someone else's documentation.
Weights arrive through the library, not a URL
Model files follow a different path entirely, and the mechanism is the Python package rather than a link. The example is three lines long:
from ultralytics import YOLO
model = YOLO("yolo26n.pt")
results = model("path/to/image.jpg")The libraries download supported pretrained models from this repository automatically whenever the requested model is not found locally, so the file name in the constructor is the whole interface. Nobody sees a URL, nobody can pin a commit, and a missing weight surfaces as a download rather than as a traceback about a path. That is the right ergonomics for a default model and the wrong ones for reproducible inference, where two runs a month apart can quietly resolve to different files. Dataset artifacts sit on a third channel again, published through repository releases and referenced from the datasets documentation. The practical consequence of three channels is that there is no single place to look when something fails to load. A missing logo is a 404 on a branch path, a missing weight is a download inside a constructor call, and a missing dataset artifact is an absent release asset, and each failure surfaces in a different subsystem with a different message.
The layout stops at yolov8 while the releases say YOLO26
The repository layout groups media by family, and the families stop where you would expect them not to. Directories exist for `yolo/`, `yolov3/`, `yolov5/` and `yolov8/`, holding banners, diagrams and example assets. There is no directory for YOLO11 or YOLO26, the two families the recent releases announce. The release numbering explains why: v8.2.0 covered YOLOv8-World and YOLOv9-C/E models, v8.3.0 was the YOLO11 release, and v8.4.0 was the YOLO26 release. The version numbers track model generations rather than the assets, so the repository's own layout lags its own changelog. The dates are uneven too, with v8.2.0 on 2024-04-17, v8.3.0 on 2024-09-29, v8.4.0 on 2026-01-13, and a push on 2026-09-14 with nothing tagged since.
AGPL-3.0 with a commercial alternative, applied to media
The licence section offers two routes rather than one. The default is AGPL-3.0, described as an OSI-approved open-source licence for collaboration and knowledge sharing, with details in the `LICENSE` file. Alongside it sits an Enterprise License, a commercial option for integrating Ultralytics software and AI models into commercial products without the AGPL-3.0 requirements, obtained by contacting Ultralytics Licensing. Applying that pairing to a media repository is worth pausing on. A logo, a favicon and a documentation screenshot are not software, so the practical question is which terms attach to a brand asset reused in someone else's product, and the page does not separate the two cases. The `documents/` directory at least keeps the Contributor License Agreement and other policy files versioned alongside everything else.
A format workflow guards files that are not code
The one badge at the top of the page points at a workflow named for formatting rather than for tests, which tells you what CI is actually doing here. There is nothing to compile in a directory of SVGs and PNGs, so the checks are about file hygiene: consistent formatting, valid metadata, nothing corrupted. The contribution guidance matches that framing, asking for changes that improve asset quality, accuracy, organization or documentation rather than code. Two things live off this repository. The Contributing Guide is on the documentation site, not in the tree, so the rules for adding an asset are not versioned with the assets. And product feedback is routed to a survey rather than to the issue tracker, leaving issues for bugs and requests about the files themselves.
Agent instruction files sit beside fourteen media directories
The top level is short and legible. Two readmes, English and Simplified Chinese, sit with a `.gitignore`, the `LICENSE` file, an `AGENTS.md`, a `CLAUDE.md`, a `.github/` directory and fourteen asset directories: `app/`, `blog/`, `docs/`, `documents/`, `im/`, `logo/`, `mkdocs/`, `og/`, `partners/`, `social/`, `yolo/`, `yolov3/`, `yolov5/` and `yolov8/`. Two entries are worth pausing on. Agent instruction files in a media repository mean the conventions for adding a logo are written for tools rather than for contributors, which is a reasonable choice for automated asset updates and a poor one for a person opening a pull request. And `im/`, holding sample images used in inference and documentation examples, is the only directory with a functional role, since those files are inputs to something rather than decoration. The remaining directories are all presentational and all named after where they are published rather than after what they contain: `mkdocs/` feeds documentation builds, `og/` produces link previews, `social/` carries icons reused across the other repositories, and `app/` holds store badges and onboarding visuals for the mobile applications. `blog/` and `partners/` complete the set. Naming by destination keeps the referring code simple, at the cost of an asset that serves two audiences having to live in one of them.
Editorial conclusion
The repository fits exactly one job, which is giving the Ultralytics ecosystem stable places to point at, and it does that job without becoming a package you install. It does not fit anyone who needs reproducible documentation builds, because every documented image URL tracks the default branch rather than a tag, or anyone treating it as a general asset bucket, since the layout is organised around one product's media. Before embedding anything from it, pin to a commit rather than following the documented `main` pattern, and check whether the file you want is a branch asset or a release artifact, because those are served from different places. Dataset artifacts additionally carry their own README, licence and usage notes that you are expected to read. The newest release is v8.4.0 from 2026-01-13 and the last push is dated 2026-09-14.
Frequently asked questions
How do I link to an image hosted in ultralytics/assets?
Use the raw GitHub URL form pointing at `main`, for example `https://raw.githubusercontent.com/ultralytics/assets/main/logo/Ultralytics_Logotype_Original.svg`. That tracks the default branch, so the referenced file can change without a version bump.
Where does ultralytics/assets store YOLO model weights?
In the same repository as the logos and media. Ultralytics libraries download supported pretrained models automatically when the requested file is not found locally, so `YOLO("yolo26n.pt")` fetches it without any URL.
How are ultralytics dataset artifacts distributed?
Through repository releases rather than the default branch, and they are referenced from the Ultralytics datasets documentation. Each one has its own README, licence and usage notes to review before use.
Which model families have directories in ultralytics/assets?
`yolo/`, `yolov3/`, `yolov5/` and `yolov8/`. There is no directory for YOLO11 or YOLO26, even though releases v8.3.0 and v8.4.0 announced those families.
Can I use ultralytics assets commercially?
The default terms are AGPL-3.0, and an Enterprise License is offered for integrating Ultralytics software and AI models into commercial products without the AGPL-3.0 requirements. The page does not separate the terms for media assets from those for models.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/ultralytics-assets)Star history chart for your README
[](https://hysenlabs.com/projects/ultralytics-assets)Was this useful?